Beruflich Dokumente
Kultur Dokumente
Abstract—this paper addresses on design of an optimal these cases: the empirically methods, directly parameters
proportional integral derivative (PID) controller by applying optimization methods, intelligent methods, and classic LQR
Lyapunov approach. The optimal PID controller is acquired methods. In this paper, a novel systematic method is presented
through minimizing an augmented integral squared error (AISE) to design an optimal PID controller. Based on Lyapunov
performance index that contains control error and at least first theory, an optimal PID controller is proposed by minimizing
order error derivative, and even may contain n orders error
the augmented integral squared error (AISE) performance
derivative. The order of error derivative is determined by the
controlled plant’s model order. The optimal control problem is index. The proposed method is unlike the classic optimal
transformed into a nonlinear constraint optimization (NLCO) control method that needs solving a riccati equation (RE). The
problem. The optimal parameters are obtained from solving the optimal PID parameters are obtained form solving a nonlinear
NLCO problem. The performances of different tuning methods constraint optimization problem. The proposed method is
are compared. The effects of control weight on optimal controller convenient to design an optimal controller that can provide
and control system’s performance are analyzed as well. The high performance for control system. The paper is arranged as:
computer simulation results are presented to demonstrate the the preliminary and optimal problem statements are presented
effectiveness of the proposed method and optimal PID controller. in Section.2. In Section.3, the systematic method of design the
optimal controller is proposed. The simulation results are
Keywords- optimal PID controller; Lyapunov approach; NLCO presents and the effects of weight factor on the control system
problem; control weight; are analyzed in Section.4. In Section.5, the study contents are
reviewed and conclusions are presented.
1. Introduction
2. Preliminary and optimal problem statements
The PID controller is widely applied in various industrial Preliminary:
control processes. Traditional PID controllers have been tuned A method is proposed to design the optimal PID controller
empirically by the methods that are proposed by Ziegler- based on Lyapunov theory. Following theorems are utilized to
Nichols and Cohen-Coon in [3-5]. Although the Ziegler- design the optimal PID controller.
Nichols (Z-N) methods are presented in 1940s, they are still Theorem.1[1,2]:
broadly used in the industrial control process. Z-N methods Sufficient and necessary conditions for the linear time
contain two tuning methods that are presented in [3-4]. One is invariant system x = Ax is asymptotical stability in a large
known as the continuous cycling (C-C) method (first Z-N scope is that: for any given Q=QT>0 there exists a matrix
method), the other is process reaction curve (PRC) method P=PT>0 that satisfies the Lyapunov algebraic equation (LAE):
(second Z-N method). The PRC method is special proposed A P + PA = −Q .
T
for the plants that contain pure time delay. Cohen and Coon
present a design method for a first order plus time delay Theorem.2[1,2]:
(FOPTD) plant in [5]. These empirically methods sometimes The linear time invariant system x = Ax , and x(0)=x0 is
bring about oscillatory responses. In recent years, many asymptotical stability, then the performance index J has the
methods have been proposed to design the optimal PID following relationships
∞
controller. An optimal PID tuning method is presented by J = x Qxdt = x (0) Px (0), A P + PA = −Q
T T T
index for FOPTD and second order plus time delay (SOPTD) Optimal problem statements:
plants in [6]. The intelligent algorithms such as immune The general transfer function Gp(s) represents the dynamic
algorithm (IA), modified genetic algorithm (MGA), modified model of linear controlled plants.
ant colony algorithm (MACA), and particle swarm algorithm b1s w + b2 s w−1 + " + bw s + bw+1 N ( s )
G p (s) = = (1)
(PSA) are employed to design the optimal PID controller in s n + a1s n −1 + " + an −1s + an M ( s)
[7-13]. Direct parameters optimization method is used to Where b1, b2 . . . bw+1 and a1 … an are the system parameters.
design the optimal PID controller in [19-22]. The classic linear Denominator’s and numerator’s order of transfer function are
quadratic regulator (LQR) methods are also employed to deg (M(s)) = n and deg (N(s)) = w respectively. For most of
design the optimal PID controller in [23-25]. To improve the practical control system n>w, so the number of poles n is large
control system’s robustness, a robust optimal PID controller is than the number of zeros w. It is noted that for an all-pole
designed in [26-27]. Most of the methods are proposed for plant b1, b2 … bw=0, additionally for a Type 1 plant an=0, and
first-order plant and second-order plant, and some of methods for a Type 2 plant an-1= an=0, and so on. The control system
are complex for design of the PID controller. The methods that diagram is shown in Fig.1. Where Gc(s) is the transfer function
are used to design PID controller in [3-27] are divided into of the PID controller, e(t) is control error, r(t) and y(t) are the
° i =0
³ 0
°
® subject to ( s.t ) : (8)
°
°(1) x = Ax
°(2) Re ( eig ( A ) ) E 0
°
°(3) m4 = f1 ( m1 , m2 , m3 ) ,! , mw+ 3 = f w ( m1 , m2 , m3 )
¯
Where Q is a control weight matrix and at least is a positive
When the command signal r(t) is a constant signal or semi-definite (PSD) real symmetric matrix. Constraint.2 of the
command signal r(t)=0, even the piecewise constant signal. equation.8 certifies the stability of the control system. The
The system’s error differential equation can be got by inverse proposed method assumes the linear controlled plant is
Laplace transform of equation.4. Hence the error differential controllable. The optimal PID controller is depended on the
equation yields parameters mi (i=1, 2,…,w+3) that are the decision variables.
( )
δ M (δ ) + N (δ ) ( k pδ + ki + kd δ 2 ) e(t ) = 0 (5) But in these decision variables, only three variables are the
independent variable because the rest variables are the
Where į is differential operator. Considering the polynomial
interrelated variable. The linear functions f1,…,fw of the
differential operator M(į) and N(į), then the error differential
constraint.3 reflect the decision variables’ inner relationships.
is acquired
3. Optimal controller design
e(t )( n +1) + a1e(t )( n ) + a2 e(t )( n −1) + " + an − w− 2e(t )( w+3) Not losing the generality, a standard third-order controlled
+ ( an − w−1 + b1kd ) e(t )( w+ 2) + ( an − w + b1k p + b2 kd ) e(t )( w+1) plant is considered in the unity feedback control system. The
plant’s transfer function and PID controller has the following
w −1
+ ¦ ( an − w+ j + b j +1k p + b j ki + b j + 2 kd )e(t )( w+1− j ) form respectively
j =1 b1s 2 + b2 s + b3 k
G p ( s) = , Gc ( s) = k p + i +kd s (9)
+ ( an + bw+1k p + bw ki ) e(t ) + bw+1ki e(t ) = 0 (6) s + a1s + a2 s + a3
3 2
s
The open looped transfer function of control system yields
The state-space model of differential equation.6 is obtained
x = Ax (7) G ( s) = Gc (s)G p (s) =
( k s + k + k s )( b s + b s + b )
p i d
2
1
2
2 3
(10)
s(s + a s + a s + a )
3 2
Where x is state vector, A is state matrix, C1×n is a parameter 1 2 3
m j + 2 = ( an − w+ j + b j +1k p + b j ki + b j + 2 kd ) ( j = 1, 2,...., w − 1)
middle decision variables and mi1 =mi+1/m1 (i=1, 2, 3, 4).
x = Ax (12)
mw+ 2 = ( an + bw+1k p + bw ki ) , mw+3 = bw+1ki § O3×1 I 3×3 ·
( )
x = e(t ) e(t ) e(t ) e(t ) (3) , A = ¨
− m14 −C1×3 ¹
1
(
¸ , C1×3 = − m3 − m2 − m1
1 1
)
To assess an optimal controller, we should pose a performance ©
index. The performance index such as the integral squared The proposed optimal control problem is formulated
error (ISE), IAE, integral time absolute error (ITAE) and
integral time squared error (ITSE) are often used for criteria.
Precision and steady-state characteristic of control system are
J * = min J ( m ,! , m ) = min ∞ xT (t )Qx(t )dt variables and three constraint conditions. In fact, the NLCO
° Gc*
1 5 ³0 problem.16 only has three independent variables. The NLCO
° 3 problem.16 can be solved by different optimization methods.
° = min ¦ qi +1,i +1 ( e(t )(i ) ) 2 dt
∞
° i =0
³ 0 Many optimization methods such as Newton's method, Quasi-
° Newton methods, Lagrange method, Conjugate Gradient
® s.t : (13)
° method, Interior Point method, and intelligent algorithms have
°(1) x = Ax been well established to solve the NLCO problem. Thus the
°(2) Re ( eig ( A ) ) E 0 optimal parameters mi* (i=1,…,5) are acquired, considering
°
°(3) m4 = f1 ( m1 , m2 , m3 ) , m5 = f 2 ( m1 , m2 , m3 ) the defined parameters in Equ.11, finally the optimal PID
¯ controller is obtained
If the optimal control problem is solved, then corresponding ki* m* − a − b k * m* m* − 1
optimal PID controller is obtained. The system characteristic Gc* = k *p + + kd* s, k *p = 4 3 2 i , ki* = 5 , kd* = 1 (17)
s b3 b3 b1
equation (CE) is D ( s ) = s 4 + m11s 3 + m12 s 2 + m31 s + m14 = 0 . Based
We should be noted that: (1) the proposed systematic method
on linear time invariant (LTI) stability theory, all the poles of can be applied in first-order, second-order, and higher-order
the CE should have negative real part that guarantees the plant as well; (2) The design method is proposed for unit
control system is asymptotic stability in a large scope. Hence, feedback control system. If the control system is not unit
the middle decision parameters mi1 (i=1, 2, 3, 4) should satisfy feedback control system, then this system can be transformed
the inequality.14 that is obtained from routh-hurwitz criterion into standard unit feedback control system by employing
(RHC). transfer function equivalent principle; (3) Time delay can be
°m11 ; 0, m14 ; 0, m11m12 − m31 ; 0 approximated by padé approximation. Thus, the presented
® 1 1 1 (14) method is also applied in a plant containing time delay.
°̄m1 m2 m3 − (m3 ) − (m1 ) m4 ; 0
1 2 1 2 1
4. Simulation results
Because the control system and its error autonomous system To prove the proposed method, let us consider following
are asymptotic stability, according to Lyapunov theorem.1-2, two controlled processes GP1(s) and GP2(s) that are the transfer
the performance index of the optimal control problem.13 is function of plant.1[7] and plant.2[6] respectively. GP1(s) is a
equal to the following performance index.15. third-order all-pole controlled object, whereas GP2(s) is a
∞
J * = min
*
J ( m1 ,! , m5 ) = min ³ xT Qxdt = min xT (0) Px(0) (15) FOPTD plant.
Gc 0
15 4e−10 s
Where P is a positive definite (PD) real symmetric matrix and G p1 ( s ) = 2 , G p 2 (s) =
P meets the LAE: AT P + PA = −Q .The optimal performance ( s + 0.9 s + 5)( s + 3) 10s + 1
The Table.1 demonstrates the optimal PID controllers and
index.15 is determined by the matrix P and initial state. Thus,
system performances in various control weight matrixes. The
the optimal control problem.13 is equally transformed by
effects of control weight matrix on the system performances
employing the Lyapunov theorem.1-2. The optimal control
and the corresponding step responses of the plant.1 are shown
problem.13 is already equally transformed into optimization
in Table.1 and Fig.2 respectively. The Fig.2 displays the
problem.16. Hence, designing an optimal PID controller is
effects of control weight on control system more intuitively.
depicted by optimization problem.16
Table.1 and Fig.2 both reflect that control weight matrix has
J * = min J ( m1 ,! , m5 ) = min xT (0) Px (0)
* large impact on the control system, particular the control
° Gc
system’s response speed and transient performances are
° s.t : mainly affected by control weight factor q1. The overshoot
°
°(1) AT P + PA = −Q increases form 6.69% to 8.47% when the weight factor q1
® (16) increases from 5 to 10, which is shown in Table.1. Fig.2 and
°(2) m11 ; 0, m41 ; 0, m11m21 − m31 ; 0, Table.1 both show that the response speed will heighten when
° 1 1 1 increasing weight factor q1. Through the study, it finds that
° m1 m2 m3 − ( m3 ) − ( m1 ) m4 ; 0
1 2 1 2 1